Neural Machine Translation (NMT) is awesome. Except when it's not. In this talk I'll review my prior work on how often NMT outputs potentially misleading translations (i.e. translations that are fluent but not adequate) and how these errors affect monolingual user trust in an MT system. I will then introduce one possible aid in helping users recognize and recover from these errors and how I plan to test the effectiveness of this approach (feedback welcome!).
Marianna J. Martindale is a PhD candidate in Information Studies at the University of Maryland. She received her BS in Computer science from BYU and MS in Linguistics (Computational) from Georgetown University. Since 2003, she has worked for the US Government supporting, building, and deploying machine translation systems in the "real world." Her research focuses on machine translation reliability and non-translator users of MT.